RepliCNN: High-resolution inference of the DNA replication program from strand-specific 3′ DNA end sequencing

During S phase, the genome is replicated in a tightly regulated spatiotemporal order described as DNA replication timing (RT). Discontinuous lagging-strand synthesis produces Okazaki fragments whose strand-specific distribution reflects replication dynamics. Here, we present RepliCNN, a deep learning framework based on one-dimensional convolutional neural networks to predict RT from Okazaki fragment distributions obtained from strand-specific 3′ DNA end sequencing methods such as GLOE-Seq, TrAEL-seq, or OK-Seq. RepliCNN also automatically annotates replication origins, termination zones, replication fork directionality, and origin efficiency genome-wide from a single dataset. Benchmarking on public and in-house human and yeast datasets using leave-one-chromosome-out cross-validation demonstrates high predictive accuracy in both wild-type and perturbation experiments, enabling comprehensive analyses of replication dynamics from strand-specific DNA 3′ end sequencing data. Highlights RepliCNN enables integrated analysis of replication timing, fork directionality and replication features from strand-specific 3′ DNA end sequencing data. High-resolution replication dynamics can be inferred from a single experiment, bypassing complex multi-fraction labelling approaches. The framework generalizes across experimental protocols, datasets, and species. This enables cost-effective comparative analysis of replication programs across biological conditions.